Efficient Community Detection Algorithm with Label Propagation using Node Importance and Link Weight

Abstract

Community detection is a principle tool for analysing and studying of a network structure. Label Propagation Algorithm (LPA) is a simple and fast community detection algorithm which is not accurate enough because of its randomness. However, some advanced versions of LPA have been presented in recent years, but their accuracy need to be improved. In this paper, an improved version of label propagation algorithm for community detection called WILPAS is presented. The proposed algorithm for community detection considers both nodes and links important. WILPAS is a parameter-free algorithm and so requires no prior knowledge. Experiments and benchmarks demonstrate that WILPAS is a pretty fast algorithm and outperforms other representative methods in community detection on both synthetic and real-world networks. More specifically, experiments show that the proposed method can detect the true community structure of real-world networks with higher accuracy than other representative label propagation-based algorithms. Finally, experimental results on the networks with millions of links reveal that the proposed algorithm preserve nearly linear time complexity of traditional LPA. Therefore, the proposed algorithm can efficiently detect communities of large-scale social networks.